Mar 3, 2025 · 1h 2m · news
Mike Krieger, Instagram CoFounder & Anthropic CPO: Where Will Value Be Created in an AI World?|E1265 · 20VC with Harry Stebbings
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
In this interview, Instagram co-founder and Anthropic CPO Mike Krieger discusses the strategic landscape of artificial intelligence, sharing insights on value creation, product design for non-deterministic systems, and the competitive dynamics between global AI labs. Krieger outlines Anthropic's transition from model provider to application builder and details the operational practices required to maintain shipping velocity in a rapidly evolving market.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Harry holds 13.6% of the talking time here. How this is scored →
Harry as informed peer 3.6 Guest teaching 3.7 Guest disagreement 1.1 Harry pushing back 3.1
speaking balance: gold is Harry, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 28:43 Mike rejects international model distillation
Mike takes a firm stance against cross-border distillation, explicitly rejecting the idea that nations should distill models from rival labs and pointing to national security and TOS concerns.
Hardest push from Harry ▶ 52:25 Harry asks if Western AI labs are over-capitalizedHarry directly challenges Mike by asking whether Western AI labs like OpenAI and Anthropic simply have too much money compared to resource-constrained innovators like DeepSeek.
Biggest teaching moment ▶ 15:57 Mike explains HCI leaky abstractions in current AI UXMike educates Harry on Human-Computer Interaction principles, explaining why picking models, managing context turns, and prompt engineering are flawed 'leaky abstractions' that must be eliminated.
Harry holds his own ▶ 39:09 Harry pushes back on generalist vs vertical consumer appsHarry counters Mike's stance against vertical applications by citing concrete everyday use cases like translation for casual users, prompting Mike to admit they reached a synthesis of views.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Harry as informed peer | Guest teaching | Guest disagreement | Harry pushing back | Why |
|---|---|---|---|---|---|---|
| Where Will Value Be Generated in the AI Decade? | 3 | 3 | 1 | 3 | Harry opens with a challenging macro VC question regarding value creation in AI, pushing on whether incumbents or startups benefit more. Mike explains the necessity of specialized data and industry-specific distribution moats. | |
| Should Startups Build for Today's Models or Future Capabilities? | 2 | 3 | 1 | 2 | Harry inquires about startup product strategy amidst rapid model shifts. Mike uses Cursor as an example of founders iterating endlessly until underlying model capabilities catch up. | |
| Is There Long-Term Value in the Foundational Model Layer? | 3 | 4 | 1 | 3 | Harry questions whether foundation model providers can capture durable value if data isn't differentiated. Mike details three sources of value: talent density, focused model traits, and high-touch enterprise AI partnerships. | |
| The Biggest Blockers to AI Progress: Evaluations and Real-World Environments | 4 | 4 | 1 | 2 | Harry cites perspectives from Alex Wang and Jonathan Ross regarding AI bottlenecks. Mike reframes the primary bottleneck away from raw compute/data to creating evals that mirror complex real-world work environments. | |
| The Future of Data: Human vs. Synthetic and Evaluating 'Vibes' | 3 | 4 | 1 | 2 | Harry references Adarsh from McCall to prompt a debate on human versus synthetic data. Mike explains why synthetic data generation requires structured environments and highlights the difficulty of evaluating model vibes. | |
| AI's Leaky Abstractions and the Future of Model Selection | 4 | 4 | 1 | 4 | Harry proposes that user model selection will disappear within 3 to 5 years like choosing search engines. Mike agrees and applies the HCI concept of leaky abstractions to model picking, chat context resets, and prompt engineering. | |
| Designing for Non-Deterministic Systems: Model Quality vs. UX | 2 | 4 | 0 | 1 | Harry asks how product leaders balance model capability against UX design. Mike contrasts Instagram's deterministic pixel reviews with building scaffolds around non-deterministic AI outputs. | |
| The Product Leader's Dilemma: Shipping Speed vs. Stability at Scale | 4 | 3 | 1 | 2 | Harry brings up a quote from Sam Altman regarding the burden of release scale. Mike elaborates on managing distinct shipping speeds across API, consumer, and enterprise customer tiers. | |
| The AI Product Marketing 'Crossy Road': Managing Rapid-Fire Releases | 3 | 3 | 0 | 3 | Harry asks if rapid-fire model launches are creating a product marketing nightmare and market apathy. Mike likens current AI PR scheduling to Crossy Road and describes locking Sonnet 3.7 announcements late on Sunday night. | |
| Navigating the 'It's So Over, We're So Back' AI Hype Cycle | 3 | 3 | 1 | 3 | Harry candidly asks whether competitor model drops trigger panic or celebration inside Anthropic. Mike discusses surviving the AI hype cycle and maintaining perspective beyond weekly benchmark swings. | |
| The Brand Differentiation of AI: Personalities, Formats, and Vibes | 4 | 4 | 2 | 4 | Harry asserts that brand identity forms the primary moat for AI models. Mike shares his social media framework of Format, Audience, and Vibes, while pushing back against cross-border model distillation. | |
| The Data Moat: Does Llama and Gemini Prove the Value is in Data? | 4 | 4 | 1 | 4 | Harry asks if Meta giving away Llama proves models have no stand-alone value without data, and cites Alex Wang on China's AI capability. Mike analyzes Gemini's video data advantage and warns against underestimating Chinese tech talent. | |
| DeepSeek's Product Impact on Anthropic: Storytelling and Velocity | 4 | 4 | 2 | 5 | Harry presses Mike on why DeepSeek achieved a cultural breakthrough that Claude failed to match. Mike acknowledges that Anthropic needs to tell its efficiency story better and move faster on raw product ideas. | |
| From Model Provider to Application Provider: Anthropic's Product Strategy | 5 | 3 | 2 | 6 | Harry directly challenges Mike's assertion that Anthropic won't build vertical apps by citing horizontal domains like translation and casual user scenarios. Mike concedes that consumer workflows reach product maturity quickly. | |
| Claude Code and the Agentic Future of Software Development | 3 | 4 | 1 | 2 | Harry asks about Anthropic's developer roadmap and potential IDE builds. Mike explains why Claude Code focuses on CLI agentic tasks rather than competing head-on with established IDEs like Cursor. | |
| The Role of the Software Developer in 3 to 5 Years | 3 | 4 | 1 | 2 | Harry asks how developer roles will evolve in 3 to 5 years. Mike envisions software engineers transitioning from manual coders to multidisciplinary reviewers and delegators operating agentic pipelines. | |
| AI Plates and Human Constraints: Alignment and Product Strategy | 4 | 4 | 1 | 3 | Harry asks if AI development velocity will hit a plateau. Mike highlights human organizational alignment as the true constraint, and admits Anthropic previously under-invested in first-party product speed. | |
| Increasing Shipping Velocity: Reclaiming the Startup Playbook | 5 | 4 | 2 | 6 | Harry bluntly asks if Western AI labs like Anthropic and OpenAI have too much money compared to lean rivals. Mike candidly agrees that product adoption previously ran ahead of true product-market fit. | |
| Rebuilding the Anthropic Product Stack from Scratch | 4 | 4 | 1 | 4 | In a quickfire round, Harry asks how much being late to first-party consumer products hurt Anthropic. Mike admits it cost them significant narrative control and momentum. | |
| The Challenge of AI Discernment and Information Privacy | 4 | 4 | 1 | 3 | Harry asks about unaddressed technical risks and queries whether Mike fears his daughter forming primary relationships with AI agents. Mike details the missing model capability of situational discernment. | |
| Europe's Regulatory Role and Entrepreneurial Future in AI | 4 | 3 | 0 | 2 | Harry asks about Europe's role in AI and quotes Dario Amodei on extending human lifespan to 150, mentioning his mother's battle with MS. Mike shares strong optimism regarding AI-driven drug discovery acceleration. |
Statements from this episode (54)
Krieger: AI foundation models will differentiate over time, not converge
“I think models over time get more different rather than more similar.”
Krieger: AI is not yet indispensable for most workers
“I still think we are in, like, day one around, is AI an indispensable part of most people's work? And I think the answer is no.”
Krieger: DeepSeek's cutting-edge AI capabilities should not surprise observers
“I think the DeepSync piece, people seem surprised that there were cutting-edge research teams there, and if you were paying attention, that part should not have been the surprising piece.”
Krieger: Anthropic under-invested in first-party product iteration and API
“I think we've, if anything, under-invested a bit in two things. One is just Having a faster iteration speed on first-party products, and then on the second part on the API side.”
Krieger: AI Startups Need Domain Knowledge and Proprietary Data for Long-Term Value
“My sense of where it ends up being most valuable to exist is places where you have some differentiated go to market, some differentiated knowledge of some particular industry or some special data that only you have access to ideally two or even three of those …”
Krieger: AI product design must build for model capabilities 3 months out
“The very thing about AI and product design is you have to dance this very delicate dance of showing the future and dreaming up what the models are currently capable at their edges, you know, cause you want to design for where they'll be, gosh, three months fro…”
Krieger: AI startups can overpromise more than incumbents due to forgiving early adopters
“And now if you're a startup, you can do a little bit more of the over promising because people are kicking your tires, the early adopters, they have a little bit more of that. Sort of willingness to engage.”
Krieger: AI breakthroughs benefit pre-existing builders over new entrants
“Often the companies that do benefit from those model generation shifts are not the ones that suddenly start that day. Like, gosh, you know, it sounds like cloud three, seven Sonic can do that. It's the ones that have been beating against the wall.”
Krieger: DeepSeek release had almost no impact on Anthropic's go-to-market
“I got this question a bunch with deep seek when deep seek came out, like, all right, what does deep seek mean for you? And I think there's things that we learned from on the tech side, just looking at what they were doing, but from a go to market and place in …”
Krieger: Selling pure API token access is a terminal failure mode
“I think the more you are just, like, maybe it's good inverting that all to see, like, what the failure mode looks like. I think it is resting on your laurels or not retaining your best people. Just believing that making the models incrementally better in every…”
Krieger: Anthropic is heavily focused on AI for office professionals
“We think a lot about office professionals at Anthropic in terms of one of the use cases that, you know, is going to be potentially really multiplied by these models in the future. Nobody's really evaluating that.”
Krieger: Creating multi-step task environments is the main AI blocker
“Figuring out how we better either break that down into component parts, which is probably part of the story, but also think about it holistically is the biggest blocker to at least one slice of progress, which is how do models go from being extremely good at e…”
Krieger: Top AI models will require both human and synthetic data
“So I think it absolutely has to be a mix. And I think the best models will come from that combination of great, like for code it's, you know, being, having good foundational understanding of code and good examples, but then also being able to explore a really …”
Stebbings: Manual AI model selection will end in three to five years
“But I think when you project yourself for three to five years, you will not be selecting which model you use.”
Krieger: Current AI Product Design Suffers From Leaky Abstractions
“And the reality is the current state of most AI product design is an extraordinarily leaky abstraction.”
Krieger: AI product development is the most complex work of his career
“It's in many ways the most complex product development work I'll ever do.”
Anthropic locked Claude 3.7 Sonnet blog post hours before launch
“We were cloud three, seven sonnet. We launched on Monday and we locked the blog post for that. Sunday night at nine PM, which is not best practice from a marketing perspective. You know, we were briefing press that day on Sunday.”
Krieger: AI customers won't switch models overnight due to custom integrations
“Over time, you start learning that, people don't just deploy models. They're doing like fine tunes or they're deploying models. Plus they've done a lot of really bespoke work to make that model be great for that use case. It's not a thing that's going to switc…”
Krieger: Social networks succeed through format, audience, and vibes
“Social networks are made of format, or formats that you have in your product, audience, and vibes.”
Krieger: AI products differentiate on model personality, scaffolding, and vibes
“I don't know what that fake formula is for AI products yet, but I think it's some version of that where there's like model model personality is probably one of them. There's likely something around the scaffolding prescriptiveness of the product that you're wo…”
Krieger: AI Labs Use Internal Distillation to Reduce Latency and Costs
“Even, like, let's take within the labs, like, I assume every single one of the labs is using, like, even within themselves, like, it is very valuable to be able to take, you know, the knowledge of your highest-end model and then be able to make it higher, you …”
Krieger: Nations Should Not Distill AI Models From Other Countries
“I think the places where this gets interesting are, one, do we want any nation to be able to be able to distill models from any other ones? Like, My personal answer is no. I think that there's value in like, even like as AI gains and capabilities being really …”
Krieger: Distillation Is Unnecessary for Frontier Open-Source AI Progress
“I think the open source models, Like take Llama, for example, like they've been able to do that from their own research and perspective and data ingestion and training. And so I guess I would say distillation does not feel essential in order to unlock those th…”
Krieger: Google Gemini benefits significantly from training on YouTube video data
“It's actually clear to me that Gemini benefits from that. Like whenever they have like a good, like video understanding demo, for example, I'm like, well, I, Somebody has like probably the largest repository of video in the world and can likely train on a lot …”
Krieger: AI benchmark evals do not indicate real-world model performance
“Evals are really useful for hill climbing and for internal research, but they don't tell the story of like, is the model going to be excellent at what it needs to be excellent or deployed for, or even if it is excellent at that thing, is it only excellent at t…”
Krieger: WeChat solved technical challenges on par with Facebook's scale
“People love talking about the, like the super app and we chat, and there was some technical challenges solved by those at scale that were of the same scale of challenges that Facebook was challenged was doing.”
Krieger: Underestimating China's frontier AI capabilities is a mistake
“It was absolutely. Be a mistake to have underestimated or continue to underestimate like China's ability to both train at the frontier especially like if they get access to compute and then continue to innovate there too.”
Krieger: AI labs will increasingly obscure model chain-of-thought outputs
“More labs either choose to not show or otherwise obscure the chain of thought down the line.”
Krieger: Anthropic trained Claude 3 with a much smaller team than competitors
“The Claude three, we were training a model at the frontier that was state of the art with a team that was much, much, much smaller. Than any other lab.”
Krieger: Anthropic product team is roughly 10% of total workforce
“Cause our team for all of Anthropic being big, you know, I think we crossed a thousand people. Our product team is, you know, maybe a 10th of that.”
Krieger: Anthropic will build general-purpose software, not bespoke vertical AI
“We are going to be building things that are general purpose as a rule with maybe some specialization at the, like, user level, but not at the, I don't anticipate us building a lot of verticalized experiences that are, like, fairly bespoke to a given workflow o…”
Krieger: Descript features some of the best product design in AI
“Descript, I think Descript is some of the best product design in AI, and, like, they've clearly put so much time into the workflow.”
Krieger: Anthropic is focusing on agentic tools, not building an IDE
“So when I think about the coding space and where we can play and add value, it really is on the agentic side. It's not on the ID side.”
Krieger: AI coding models cannot run autonomously for hours without humans
“Recognize that they're not yet at the place where for many use cases, you can let them kind of run free for hours. You need that more human in the loop piece.”
Krieger: Most of Anthropic's best product ideas come from engineering prototypes
“Many, maybe even most of our good product ideas come from our engineers and come from them prototyping.”
Krieger: Software engineers will become AI delegators within three years
“How do we evolve from being mostly code writers to mostly Delegators to the models and code reviewers. That's what I think the work looks like three years from now. It's coming up with the right ideas, doing the right user interaction design, figuring out a de…”
Krieger: Startups have an AI alignment advantage over large incumbents
“It's why I'm really bullish on at least startups being able to explore the space because, you know, I remember this from my, both Instagram and artifact days, like when it's just a couple of you, like alignment is a coffee conversation in an afternoon rather t…”
Krieger: AI models are three-plus years from solving product strategy
“That, that's still a very human problem that I think we're at least three years away from the models being, being solving at that level of abstraction.”
Krieger: Internal dogfooding of Claude Code directly improved Claude 3.7 Sonnet
“As a really, you know, specific example with cloud code within, you know, a week of it being deployed internally, we had found a way in which one of the sort of tools that it has access to the model wasn't using as well as it could have, and that made its way …”
Krieger: Instagram allocated roughly 95% to product and 5% to API
“On Instagram. It was easy. It was like, 95% product, five percent API, and it was, you know, that's all we really needed to do.”
Krieger: Anthropic became overly calcified by adopting a large company playbook
“We got too calcified, I think. And like, oh, well, this is on this team's plate versus this team's plate. And oh, you can't get this done this quarter because it's not on this team.”
Krieger: Anthropic's current product adoption outpaces true product-market fit
“The adoption that we've gotten of our products is ahead of their actual like true Product market fit because they are still the best ways of getting the models. And I don't think that's durable over time.”
Krieger: OpenAI ships initial products faster than Anthropic
“They've moved faster at shipping V-ones, even ahead of where the model is sometimes.”
Krieger: OpenAI lags Anthropic in product personality and cohesion
“Probably personality and having the features they build be cohesive.”
Krieger: OpenAI effectively balances consumer product with API platform
“I think that they've balanced first party product development and an API that like people, people use at scale as well. And I think that they we had an Instagram principle that was do the simple thing first. And I think they often do the simple thing first.”
Krieger: Claude and ChatGPT were initially built only as model showcases
“Claude AI and probably ChatGPT.com were, like, very much, like, initially just built to be sort of showcases of the models and not really built in a lot of ways to be the right, like, The sort of foundational for like a much more complex sort of multi product …”
Krieger: Anthropic is actively rebuilding Claude's core user experience
“We have an active effort right now around tearing down some of that and rebuilding the core UX to just feel good. It doesn't feel great right now.”
Krieger: AI Companies Cannot Build Moats Without First-Party Products
“And I think that there's, you'll, you'll miss out and not have enough of a durable moat if you're not equally investing or maybe even investing even more on the first party side of things.”
Krieger: Being late to first-party products hurt Anthropic's narrative
“I think significantly if you take a deep seek moment, right? Like ideally that the, like the story of, oh, there's more than one sort of front or leading edge API, sorry, AI product to be used is some, a narrative that we should have captured. I think it hurt …”
Mike Krieger: AI Model Privacy Discernment Is Underappreciated and Under-Researched
“And I think models, this is very underappreciated and probably under-researched as well from like a model capabilities perspective because models fundamentally want to be helpful. And that is not always what you want them to be. And there's a safety case for t…”
Krieger agrees with Alexandr Wang: Most future friends will be AI
“I've had so many conversations with Alex Wang about this because he has this whole thing about how in the future most friends will be AI friends. And you know, I don't think he's wrong.”
Krieger: German data privacy standards shape Anthropic's product design
“Even as we think about doing our product design and data privacy and, you know, selling to German users or German companies, there's a different set of questions that get asked that are often very helpful questions.”
Krieger: Application startups can iterate faster than major AI labs
“Building applications on top of these models becomes, it is a lot easier, and you can go from zero to one, and you can be more nimble than even these labs that are gonna all have, like, you know, tens or hundreds of millions of users, and you have to move slow…”